# Path to the training dataset CSV file file_path: data/grpo/train.csv # Split of the dataset split: train # Maximum number of training samples to use max_rows: 10000 # Random seed for dataset shuffling seed: 42